the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Feb 2025
Brief Communication: Daily, gap-free snow cover information based on a combination of NPP VIIRS and MODIS data
Abstract. Combining Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) snow cover data and applying cloud- and data gap interpolation steps can be utilized to generate daily, gap-free snow cover information. Provided by the German Aerospace Center (DLR) under the name “Global SnowPack”, this product has undergone several improvements, comprising the inclusion of NPP VIIRS, a new threshold for the Normalized Difference Snow index (NDSI), and considerable validation. The Global SnowPack offers unique opportunities for analyzing time series of snow cover data since September 2000 and is freely available for visualization and download from DLRs GeoService.
- Preprint
(556 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 10 May 2025)
-
RC1: 'Comment on egusphere-2025-382', Anonymous Referee #1, 24 Mar 2025
reply
General comments
This short paper presents an upgrade of GSP, a daily global gap-free snow cover extent dataset derived from NASA MODIS and VIIRS snow products. The main difference with the previous GSP dataset (Dietz et al. 2015) lies in the integration of VNP10A1 products. As a result, the new GSP dataset has a finer resolution of 371 m instead of 500 m. The manuscript also includes an evaluation of the GSP dataset using Landsat images.
The GSP dataset offers an interesting compromise between coverage (global), spatial resolution (371 m), latency (3 days) and ease of access (very intuitive interface). The added value of VIIRS data is considerable and well highlighted in Figure 3. The convenient data formatting and encoding makes it easy to use for climate studies.
However, the article has several shortcomings.
- the introduction does not explain the stated what’s the added value of your GSP dataset with respect to other similar gap-free products (e.g. Modis gap filled M[O-Y]D10A1F) or IMS (https://doi.org/10.7265/N52R3PMC). A review of other snow cover extent products is clearly behind the scope of the paper, but there could be the room for briefly highlighting why one should prefer this dataset.
- The evaluation of the GSP was performed using in situ data and Landsat products. Regarding Landsat the only result is one sentence “The comparison with Landsat resulted in an F1 score of 0.94”. The discussion of the results is very limited. In the conclusion the authors wrote that the “uncertainties can easily be estimated and included in an uncertainty layer”, however there is no further indication on how to actually estimate this uncertainty and the product does not provide such layer. The lack of information regarding Landsat products and the minimalist description of the results could prevent potential users from using this dataset.
Minor comments
- In the absence of a user manual, more details about the dataset would be useful for users: is it operational ? What’s the latency ? In which grid is it distributed ?
- L28: An illustration (a sample) of the dataset would be a nice addition.
- L32: « Landsat data » Which Landsat ?
- L33: « A total of 381 Landsat scenes » : the geographical and temporal distribution of these tiles is briefly given later (“distributed worldwide and covering all ecozones and seasons”). It would be more straightforward to give this information in paragraph 2 (and not 2.1). Also, additional information about these tiles could be useful: are mountain regions and high latitudes covered? In which years were these images acquired? .
- L33-35: The sentence is confusing. A reader cannot understand how the reference data were prepared. In (Koeheler et al. 2022) the NDSI is defined between a blue and a NIR band, in SnowPex exercise there are few different methods used for snow cover detection. Can you clarify?
- L36: GMTED2010 is a multi resolution dataset. Which spatial resolution was used?
- L39-41: it is unnecessary to repeat here the motivation of this work especially in a “Brief communication”.
- L43: The daily combination step could potentially benefit from satellite viewing angle information. Plenty of evaluations show how MODIS/VIIRS information is more reliable closer to nadir [1], [2], [3].
- L55: What is the 3000 in the parenthesis next to MODIS products ?
- L66-77: should be in the Data & Method section.
- L70 (Figure 2 legend) : MOD10 identifier is a bit confusing here. MOD10 refers to MODIS/Terra products (e.g. MOD10A2). Please refer to the products ID (MOD10A1,MYD10A1,VNP10A1).
- L70 (Figure 2): maybe a logarithmic scale for the snow depth threshold would be more appropriate.
- L80: « This accuracy….Figure 2b). » Can you elaborate? Accuracy reaches a maximum during summer months and decreases in the winter, meaning that accuracy is lower when the snow cover is greater at global scales. Any explanation?
- L81 Again inconsistent notation MOD10/VIIRS.
- L84 (Figure 3) Can you specify for which year or years was this plot computed?
References
[1] K. R. Arsenault, P. R. Houser, et G. J. M. De Lannoy, « Evaluation of the MODIS snow cover fraction product », Hydrological Processes, vol. 28, no 3, p. 980‑998, 2014, doi: 10.1002/hyp.9636.
[2] K. Rittger, K. J. Bormann, E. H. Bair, J. Dozier, et T. H. Painter, « Evaluation of VIIRS and MODIS Snow Cover Fraction in High-Mountain Asia Using Landsat 8 OLI », Frontiers in Remote Sensing, vol. 2, 2021,
[3] K. Rittger, K. J. Bormann, E. H. Bair, J. Dozier, et T. H. Painter, « Evaluation of VIIRS and MODIS Snow Cover Fraction in High-Mountain Asia Using Landsat 8 OLI », Frontiers in Remote Sensing, vol. 2, 2021,
Citation: https://doi.org/10.5194/egusphere-2025-382-RC1 -
AC1: 'Reply on RC1', Andreas Dietz, 04 Apr 2025
reply
We appreciate very much the time and effort that the reviewer took to review our manuscript and provide us such valuable feedback. Before addressing the comments in detail, I want to just mention that because of the “Brief Communication” format, we were very limited in terms of space and number of figures/tables.
To answer the two major comments:
1. the introduction does not explain the stated what’s the added value of your GSP dataset with respect to other similar gap-free products (e.g. Modis gap filled M[O-Y]D10A1F) or IMS (https://doi.org/10.7265/N52R3PMC). A review of other snow cover extent products is clearly behind the scope of the paper, but there could be the room for briefly highlighting why one should prefer this dataset.
Our answer: Compared to the NSIDC products MOD10A1F and MYD10A1F, several steps are taken to fill gaps. Instead of only looking at the preceding days, a 3-day interpolation (in both directions) and a topographic interpolation are also performed. Furthermore, the information from three sensors (Terra and Aqua MODIS and NPP VIIRS) is initially combined daily, which fills most of the gaps. The Interactive Multisensor Snow and Ice Mapping System (IMS) also offers a combination of different sensors, but not at this resolution (IMS has a 1 km resolution) and not globally (IMS only covers the Northern Hemisphere). We will make sure to add this information to the manuscript to make it clearer to the reader.
2. The evaluation of the GSP was performed using in situ data and Landsat products. Regarding Landsat the only result is one sentence “The comparison with Landsat resulted in an F1 score of 0.94”. The discussion of the results is very limited. In the conclusion the authors wrote that the “uncertainties can easily be estimated and included in an uncertainty layer”, however there is no further indication on how to actually estimate this uncertainty and the product does not provide such layer. The lack of information regarding Landsat products and the minimalist description of the results could prevent potential users from using this dataset.
Our Answer: Again, the reason we only included this one, very short statement about the Landsat validation is the limited space we had with only 4 pages. After discussing this issue with the editor, we will extend the manuscript now slightly to include more information about the Landsat comparison, the location of the test sites, and details about the accuracy assessment. The quality layer is created during the process of cloud interpolation. Depending on the method that is used to fill the gap, and the duration of the cloud cover in every single pixel location, the quality of that pixel can be calculated. We conducted extensive tests and simulations of artificial cloud cover into clear scenes, interpolating them, and determining the uncertainty connected to the different interpolation steps and cloud cover durations. These simulations help us to construct a quality layer which is provided together with our Global SnowPack dataset, and a short description about that will be included in the revised manuscript. Here is also a figure illustrating the location of the Landsat validation sites (quality is not the best due to restrictions uploading figures in the answers, but hopefully enough to get an idea):
Minor comments and our answers:
- In the absence of a user manual, more details about the dataset would be useful for users: is it operational ? What’s the latency ? In which grid is it distributed ?
Answer: It is operational, provided daily with a 3-4 day time lag to Real time conditions. The grid is WGS 84 (EPSG: 4326), covering the entire globe. The spatial resolution is 12 arc-seconds (approximately 370 meters)
- L28: An illustration (a sample) of the dataset would be a nice addition.
Answer: We unfortunately cannot add any more figures, but the latest product is always available under the following link: https://geoservice.dlr.de/web/maps/eoc:gsp:daily
- L32: « Landsat data » Which Landsat ?
Answer: All available Landsat 5, 7, 8, covering all continents. There will be a more detailed description about that in the revised manuscript
- L33: « A total of 381 Landsat scenes » : the geographical and temporal distribution of these tiles is briefly given later (“distributed worldwide and covering all ecozones and seasons”). It would be more straightforward to give this information in paragraph 2 (and not 2.1). Also, additional information about these tiles could be useful: are mountain regions and high latitudes covered? In which years were these images acquired? .
Answer: The Landsat tiles cover all regions, also mountains worldwide. The whole Landsat part will be revised in the manuscript to make sure we incorporate all these comments.
- L33-35: The sentence is confusing. A reader cannot understand how the reference data were prepared. In (Koeheler et al. 2022) the NDSI is defined between a blue and a NIR band, in SnowPex exercise there are few different methods used for snow cover detection. Can you clarify?
Answer: The full descritpion of how the Landsat data was processed is available in Koehler et al 2022. Essentially, all methods described by SnowPex have been tested and we used the one that produces the best results for the Alps and Pyrenees Mountains. That could of course mean that in other regions, another approach would have worked slightly better, but we assume that our classification is accurate. We will add more information about this to the then extended Landsat part of the revised manuscript
- L36: GMTED2010 is a multi resolution dataset. Which spatial resolution was used?
Answer: We use the 7.5 arc-second resolution and resampled it to 12 arc-seconds using bilinear resampling.
- L39-41: it is unnecessary to repeat here the motivation of this work especially in a “Brief communication”.
Answer: Will be modified in the revised manuscript
- L43: The daily combination step could potentially benefit from satellite viewing angle information. Plenty of evaluations show how MODIS/VIIRS information is more reliable closer to nadir [1], [2], [3].
Answer: Thanks a lot for this advice. We are thinking to include this aspect in a new update. For now, if more than one observation is cloud-free, selection is pixel-based, initially on the quality flag in the "Basic_QA" layer. If the values are equally good, the prioritization is based on VNP, MYD, and then MOD – since the focus is on better spatial resolution, and MYD will be added later for better lighting (the equator crossing is at 1:30 p.m. local time).
- L55: What is the 3000 in the parenthesis next to MODIS products ?
Answer: The 3000 in the parentheses indicates that these are the most appropriate NDSI thresholds after the MOD10A1 or MYD10A1 products were upscaled from 2400 to 3000 pixel size (to match the spatial resolution of VIIRS). But since this is not relevant for the reader, we will remove it from the revised manuscript
- L66-77: should be in the Data & Method section
Answer: we will move it to the data and methods
- L70 (Figure 2 legend) : MOD10 identifier is a bit confusing here. MOD10 refers to MODIS/Terra products (e.g. MOD10A2). Please refer to the products ID (MOD10A1,MYD10A1,VNP10A1).
Answer: we will make sure to clarify this and include the full product ID
- L70 (Figure 2): maybe a logarithmic scale for the snow depth threshold would be more appropriate.
Answer: We will look into that and test whether a logarithmic visualization is better
- L80: « This accuracy….Figure 2b). » Can you elaborate? Accuracy reaches a maximum during summer months and decreases in the winter, meaning that accuracy is lower when the snow cover is greater at global scales. Any explanation?
Answer: The reason for the higher accuracy in summer is the reduced cloud cover and particularly the much smaller snow cover dynamics: Snow field and glaciers which exist in summer are much more stable, so interpolating a cloud gap – even if it persists many days – will not introduce many errors. In winter, when much more snow cover is present, more clouds are present, and the dynamics are much higher, cloud gaps and polar darkness can last many days and therefore introduce higher uncertainties. The polar night gap itself can be 90 days or more – usually not in very dynamic regions when it comes to snow cover, but still, errors happen more frequently here than in summer. In addition, lighting conditions are worse in winter, which makes snow detection more challenging.
- L81 Again inconsistent notation MOD10/VIIRS.
Answer: Will be addressed
- L84 (Figure 3) Can you specify for which year or years was this plot computed?
Answer: This figure is a compilation of all years in which MODIS and VIIRS were available (since January 19, 2012). The orange and purple parts of the figure show the information gain from the pixels added by VIIRS.
Citation: https://doi.org/10.5194/egusphere-2025-382-AC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 92 | 22 | 6 | 120 | 7 | 6 |
- HTML: 92
- PDF: 22
- XML: 6
- Total: 120
- BibTeX: 7
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 49 | 39 |
| France | 2 | 17 | 13 |
| China | 3 | 14 | 11 |
| Germany | 4 | 11 | 8 |
| Chile | 5 | 4 | 3 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 49